Information Propagation in the Flickr Social Network

1
Information Propagation inthe Flickr Social
Network
A Measurement-driven Analysis of
Meeyoung Cha
Alan Mislove Max Planck Institute for
Software Systems (MPI-SWS)
Krishna P. Gummadi
The 18th International WWW Conference
2
Information propagation in OSNs

• Online social networks (OSNs) are popular
• In 2007, 1.2 billion spent on advertisements in
  OSNs
• OSNs have potential for information propagation
• Viral marketers plan to use OSNs to spread
  content and ads
• OSNs are already being used for political
  campaigning, content sharing, product
  advertising

Are social links important in information flow?
3
This talk

• Goal characterize information propagation
  patterns in Flickr
• 1. How quickly does information spread over
  time?
• 2. How widely does information spread in the
  topology?
• Implications of our work
• 1. Modeling predict and estimate near-future
  trends
• 2. System design recommendation and content
  distribution

How does information flow in online social
networks?
4
Information flow mechanisms

• Featuring (front page, hotlists)
• External links
• Search results
• Links between content
• Word-of-mouth through online social links

5
Part1. Measurementmethodology
Part2. Spreading patternover time
Part3. Spreading patternover topology
6
Measure of photo popularity

• Possibilities
• Views
• Comments
• Favorites
• Users mark a photo as a favorite
• Represents who liked the photo
• We call these users fans

7
Data needed for this study

• 1. Photo uploader and upload time
• 2. Sequence of users favorite-marking
• 3. Social network topology

Like it
Like it
Like it
Like it
8
Gathering the data

• Crawled a substantial fraction of the Flickr
  social network
• 2.5M users and 33M friend links snowball sampled
• (crawled network forms a large weakly connected
  component)
• Repeated crawls daily for 104 days
• Gathered the list of favorite pictures for all
  users
• 34M bookmarks of 11M distinct photos
• (includes the exact time stamp of bookmarks)

9
Flickr social network

• Our sample single connected component of users
• High reciprocity (68 social links are
  bidirectional)
• Power-law node degree distribution (avg14,
  max26,342, a1.7)
• Short average path length of 5.67

Small-world properties Network structure is
promising for information spreading
10
Part1. Measurementmethodology
Part2. Spreading patternover time
Part3. Spreading patternover topology
11
Example pattern1 Steady-growth
London cycling by lomokev

• Gained new fans at a relatively constant rate

12
Example pattern2 Growth-spike
One would. by antimethod

• Sudden increase in fans over a short time period

13
Example pattern3Dormant

• Unknown to many users or stop gaining fans

Velcro being pulled apart by Trazy
Different popularity growth patterns indicate
that photos spread through different information
flow mechanisms
14
Aggregate growth pattern

• 5,346 photos (older than a year have more than
  100 fans)

Popular photos gain popularity slowly and
steadily.Flickr users take a long time to learn
about popular photos.
15
Comparison of growth patternstheory vs. practice

• Popular theories (Diffusion of innovations, Bass
  diffusion model) suggest an S-curve growth
  pattern

Theory
Contrary to popular theories, photos spread slowly
16
Part1. Measurementmethodology
Part2. Spreading patternover time
Part3. Spreading patternover topology
17
Comparison of locations users vs. fans1

• 3,685 popular photos with more than 100 fans
• Uploaders can reach a small fraction of users by
  1-2 hops

18
Comparison of locations users vs. fans2
users
fans
fans
users
users
fans
fans
users
High content locality around photo uploaders
Even popular photos do not spread widely in the
network
19
Identifying information flow through social links

• Did a particular bookmark spread through social
  links?
• No if a user bookmarks a photo when none of his
  friends bookmarked the same photo
• Yes if a user bookmarks a photo after any one of
  his friends bookmarked the same photo

20
What role do social links play?

• Based on the 104 daily crawls, 53 of all
  favorite-marks estimated to have traversed
  through social links
• Time exposed to a photo prior to favorite-marking
  with respect to the first friend who liked the
  same photo
• med60, avg140 days

Social links are crucial in information
spreading Information took a long time to spread
across each link
21
Summary

• OSNs have small-world properties potential for
  viral marketing
• To date, little is known about spreading patterns
  
• We studied temporal and spatial spreading
  patterns in Flickr
• Three observations
• 1. Information does not spread quickly in the
  network
• 2. Even popular photos show high content locality
• 3. Social links play a crucial role, but
  information flows very slowly at each hop
• Calls for further analysis

22
Data available at http//socialnetworks.mpi-sws.or
g/
Thank you